Multi-task evaluation on Average GSM8K, HumanEval, ARC-c
60.77AccuracyReMix
Evaluation Results
| Method | Links | |
|---|---|---|
| ReMixType=Mixture, Params=0.070B2026.03 | 60.77 | |
| rsLoRAType=Weight Modulation, Params=0.028B2026.03 | 57.95 | |
| MixLoRAType=Mixture, Params=0.101B2026.03 | 57.43 | |
| DoRAType=Weight Modulation, Params=0.127B2026.03 | 56.61 | |
| LoRAType=Weight Modulation, Params=0.112B2026.03 | 56.36 | |
| HydraLoRAType=Mixture, Params=0.084B2026.03 | 55.1 | |
| Few-ShotType=No Tuning, Params=N/A2026.03 | 51.66 | |
| P-TuningType=Prefix Injection, Params=0.001B2026.03 | 34.89 | |
| VB-LoRAType=Mixture, Params=0.675B2026.03 | 29.09 | |
| (IA)3Type=Weight Modulation, Params=0.001B2026.03 | 21.02 | |
| Prompt TuningType=Prefix Injection, Params=0.000B2026.03 | 18.22 | |
| Zero-ShotType=No Tuning, Params=N/A2026.03 | 13.41 | |
| Prefix TuningType=Prefix Injection, Params=0.024B2026.03 | 10.37 |